Fingerprint representation using localized texture features

Fingerprint representations can be broadly divided into three categories: image level, texture features and minutiae features. Both image based and texture based representations require accurate alignment before comparison. This presents a problem since accurate registration of fingerprints is challenging. On the other hand, minutiae based matchers are invariant to changes in orientation and position, but completely ignore the rich visual content in the image. In this paper, we present a localized texture based representation scheme that relies solely on visual content for identification and at the same time does not require absolute alignment. We outline techniques to efficiently compute these features and also propose an algorithm to perform identification based on these features. Our experimental evaluations over database of several sizes show that the proposed features are both accurate and scalable.